Abstract

In this article, I'll explain a simple but uncomfortable truth: Big AI companies are learning from your discoveries and adding them to their products. You spend hours discovering something, and a few months later the same thing becomes the platform's "new feature."

This isn't a conspiracy theory. Sam Altman says it openly. Academic research supports it. And on January 12, 2026, Anthropic's Cowork announcement demonstrated it live.

So what can a solo developer do? That's what we'll discuss.

1. How Does the Big Fish Swallow the Small Fish?

1.1 The Learning Loop: You Work, They Watch

Think about it: You're using an AI tool. Let's say ChatGPT or Claude. You spend hours discovering things. Maybe a special prompt technique. Maybe a workflow. Maybe a more efficient way to use the tool.

Question: Who sees this discovery?

Answer: You and... the platform company.

Sam Altman, CEO of OpenAI, doesn't hide this:

"It's important to release early versions and update frequently. We call this 'iterative deployment.'" [^1]

What does this mean? It means: They release the product early, watch how you use it, learn, and update. You are their free research laboratory.

Altman says it more explicitly:

"OpenAI released early versions, got rapid feedback from the real world, listened to users, developed based on what people actually wanted." [^2]

Notice: "Listened to users." That's a nice phrase. But what actually happened: They analyzed billions of prompts from millions of users. Which techniques work? Which workflows are popular? Which features are missing?

Result: You discover, they learn, then they package your discovery as a "new feature" and sell it.

1.2 A Real Example: Cowork

On January 12, 2026, Anthropic announced a new product called "Cowork." This product makes Claude usable for non-coders.

But wait. Power users were already doing this. They were setting up MCP servers. Writing custom hooks. Creating slash commands. For months, forums and Discord channels were full of these techniques.

And now? The same things are offered as "native features" in Cowork. $100-200 per month.

This is no coincidence. This is a strategy called platform envelopment. In plain terms: The platform swallows the ecosystem around it.

1.3 Information Asymmetry: Who Holds the Cards?

Imagine a poker table. On one side is you, with only your own cards. On the other side is the platform company, able to see everyone's cards.

This asymmetry grows over time. You discovered something? The platform already saw a hundred thousand people's similar discoveries. You made a mistake? The platform analyzed millions of errors.

1.4 Personalization: They Know You Better Than You Know Yourself

Sam Altman also talks about AI's personalization potential:

"AI learned my culture, values, and life by talking to me over the years." [^3]

Read that sentence again. "Learned over the years." The platform knows who Altman is, how he thinks, what he wants.

Question: How well does the platform know you?

If you've used the same service for months or years, the answer is probably "quite well." Your prompt style. Your work habits. Your frequent mistakes. All recorded.

2. Scientific Background: Why Is This Inevitable?

2.1 Scaling Laws: Does Bigger Always Win?

AI researchers studied how models develop. What they found is simple but important:

As models grow, data increases, and compute power rises—performance improves predictably. [^4]

This is a phenomenon called "scaling laws." Kaplan and colleagues demonstrated this mathematically in 2020.

What does it mean for a solo developer?

This: You can't win the compute race. Google, Microsoft, Anthropic, OpenAI—you can't compete with these companies' data centers.

2.2 Emergent Abilities: Gone Today, Here Tomorrow

Wei and colleagues discovered something interesting in 2022:

AI models suddenly gain new abilities when they reach a certain scale. [^5]

This is called "emergent abilities." The important thing: These abilities cannot be predicted. Today a model can't do something, tomorrow it can with an update.

What does it mean for a solo developer?

A technique that's your "secret weapon" today could become the platform's standard feature tomorrow. Maybe even better.

2.3 Agentic Systems: Orchestration Is Everything

Andrew Ng, one of the pioneers of AI, popularized a concept called "agentic AI." [^6] Simply:

Ng's most striking finding:

A simple model with the right orchestration can outperform a much more powerful model.

This is a critical point. Raw power isn't everything. How you use it matters.

What does it mean for a solo developer?

Here's where there's hope. You can't win the compute race, but you can win the orchestration race.

3. So What Should We Do?

3.1 Accept the Facts

First, let's accept some facts:

1. Platform advantage will grow. Scaling laws guarantee this.
2. Today's technique could be tomorrow's commodity. Emergent abilities make this possible.
3. You are their laboratory. Iterative deployment enables this.

Is this gloomy? Maybe. But we can't strategize without seeing reality.

3.2 Orchestration: The Solo Developer's Weapon

Remember Andrew Ng's finding: Orchestration can be more important than raw power.

What's the big companies' advantage? Compute power, data, number of engineers.

What could be your advantage?

1. Speed and Flexibility

Big companies move slowly. Bureaucratic processes, approval mechanisms, product managers, meetings. You can think of an idea in the morning and try it by evening.

2. Domain Knowledge

The platform is general-purpose. For everyone. You know your field, your customers, your problems. This knowledge is valuable.

3. Customization

The platform uses a "one size fits all" approach. You can customize to your own needs.

3.3 Three Axes: Privacy, Latency, Customization

Where can you compete as a solo developer?

Privacy: Your customers might care where their data goes. "Your data never leaves our servers" is a powerful value proposition.

Latency: Cloud services experience network latency. Local solutions can respond instantly.

Customization: The platform offers the same thing to everyone. You can shape it to your customer's exact needs.

It's possible to beat the platform on these three axes.

3.4 Velocity: The Only Sustainable Advantage

In the long run, the only sustainable advantage is speed.

The platform will copy your feature. This is inevitable. But by then, you may have moved three steps ahead.

This is a race. Stopping means death. But if you keep running, you can always stay ahead.

4. The Future: From Writing Code to Managing Agents

4.1 The Paradigm Shift

A trend emerging in late 2025: Software development is shifting from writing code to managing AI agents.

Before:

Developer → Writes code → Tests → Deploys

Now:

Developer → Directs agents → Agents produce code → Developer supervises

This shift is just beginning. But it's accelerating.

4.2 2026-2027 Predictions

1. "Agentic IDE" will become widespread Code editors will directly support AI agent management.

2. MCP ecosystem will explode Model Context Protocol (the standard for AI to talk to external tools) will spread. Thousands of ready-made integrations can be expected by the end of 2026.

3. "Agent Store" will emerge Just like buying apps on the App Store, you'll buy ready-made AI agents.

4. Adapt or be eliminated Developers who can't adapt to this change will struggle. Those who adapt could see 10x productivity gains.

4.3 Human-Computer Interaction Questions

This transition raises fundamental questions:

• Control: How much control should we hand over to AI?
• Transparency: Can we understand AI's decisions?
• Trust: When should we trust, when should we question?

These questions aren't answered yet. But we must find the answers.

5. Conclusion: What Should the Poor Solo Developer Do?

5.1 Brief Summary

In this article, I explained:

1. Platform envelopment is real. Big companies systematically add user innovations to their own products.

2. Science supports this. Scaling laws, emergent abilities, agentic patterns—all explain this trend.

3. Defense is possible. Orchestration, domain knowledge, speed, customization—these are the solo developer's weapons.

4. The future will be different. The shift from writing code to managing agents seems inevitable.

5.2 What to Do?

1. Accept the truth. The platform is watching you and learning. This isn't paranoia, it's the business model.

2. Move fast. Your advantage is temporary. Use it.

3. Focus on your domain. Don't compete with general-purpose platforms. Find your niche.

4. Learn orchestration. Not raw model power, but how you use it matters.

5. Be ready for change. 2026-2027 will bring a paradigm shift. Those prepared will win.

5.3 Final Words

Spinoza defined "conatus" in the 17th century: Every being's effort to preserve and strengthen itself.

The platforms' conatus is clear: Grow, swallow, dominate.

What should the solo developer's conatus be? I think: Learn fast, adapt fast, move fast.

This isn't a war. This is a dance. The platform grows, you dodge. By the time the platform catches you, you've already moved somewhere else.

Exhausting? Yes. But not impossible.

And maybe—just maybe—this dance makes you better.

References

[^1]: Altman, S. (2023). Stanford Lecture on AI Development. Marketing AI Institute.

[^2]: Fortune. (2023). "The Inside Story of ChatGPT's Development." Fortune Magazine.

[^3]: Altman, S. (2024). Big Technology Podcast Interview.

[^4]: Kaplan, J., et al. (2020). "Scaling Laws for Neural Language Models." arXiv:2001.08361.

[^5]: Wei, J., et al. (2022). "Emergent Abilities of Large Language Models." arXiv:2206.07682.

[^6]: Ng, A. (2024). "Agentic AI Design Patterns." DeepLearning.AI & Sequoia AI Ascent.

Version: 2.0.0 | Date: 2026-01-13